Past Event: CSEM Student Forum

Numerical solvers and deep learning models as web services and scalable applications

Kliment Minchev, CSEM Student, Oden Institute, UT Austin

Abstract

Software applications powered by deep learning models and deterministic solvers are becoming more ubiquitous, powerful, and relied upon. However, for researchers, it is still not easy to deploy the work to reach an evaluator or an ordinary user. Research and development work remains heavily siloed, user experience is fragmented, and cloud resources are severely underutilized. Even today, the technology stack required to launch a simple web service powered by a deep learning model or solver can be intimidating to set up, time consuming to troubleshoot, and cost-sensitive to inefficiencies.
 

This talk will review all requirements, technology stack, and process to deploy a deep learning model as a scalable web service. Several platforms are also on the rise, e.g., Streamlit and Gradio (among others), to allow quick deployment of code related to approximation or predictive models as containerized cloud applications. Such platforms contribute to the vision that quality research will accelerate with the proliferation of web services, cloud computing, and exposure to the outside world. 

Biography

Kliment Minchev is a CSEM master's student. His career has been focused around data science and developing technology products, from anomaly detection in time series data, to digital signal processing and sensor design, to most recently cloud native full-stack web development. In his spare time, he is an avid snowboarder and surfer.